Summary Stats

agr <- d %>% 
  group_by(Category,Task) %>% 
  summarize(MeanAccuracy = mean(Accuracy),
            SD = sd(Accuracy))

## `summarise()` has grouped output by 'Category'. You can override using the
## `.groups` argument.

print(agr)

## # A tibble: 6 × 4
## # Groups:   Category [3]
##   Category Task     MeanAccuracy    SD
##   <chr>    <chr>           <dbl> <dbl>
## 1 Adjs     Concrete        0.801 0.399
## 2 Adjs     Valence         0.934 0.249
## 3 Nouns    Concrete        0.897 0.304
## 4 Nouns    Valence         0.960 0.195
## 5 Verbs    Concrete        0.917 0.276
## 6 Verbs    Valence         0.956 0.205

Start with Accuracy

agr <- d %>% 
  group_by(Task) %>% 
  reframe(MeanAccuracy = mean(Accuracy), 
          CILow = ci.low(Accuracy), 
          CIHigh = ci.high(Accuracy)) %>%
  mutate(YMin = MeanAccuracy - CILow, 
         YMax = MeanAccuracy + CIHigh)
# View(agr)

dodge = position_dodge(.9)
ggplot(data=agr, aes(x=Task,y=MeanAccuracy, fill = Task)) +
  geom_bar(position=dodge,stat="identity") + 
  geom_errorbar(aes(ymin=YMin,ymax=YMax),width=.25,position=position_dodge(0.9))

  # theme(axis.text.x = element_text(angle = 45, hjust = 1)) +
  # guides(fill = "none")

agr <- d %>% 
  group_by(Task,Category) %>% 
  reframe(MeanAccuracy = mean(Accuracy), 
          CILow = ci.low(Accuracy), 
          CIHigh = ci.high(Accuracy)) %>%
  mutate(YMin = MeanAccuracy - CILow, 
         YMax = MeanAccuracy + CIHigh)
# View(agr)

dodge = position_dodge(.9)
ggplot(data=agr, aes(x=Task,y=MeanAccuracy, fill = Category)) +
  geom_bar(position=dodge,stat="identity") + 
  geom_errorbar(aes(ymin=YMin,ymax=YMax),width=.25,position=position_dodge(0.9))

# ggsave("../graphs/conc-abs_categoryXtask.pdf",width = 5, height = 3)

agr <- d %>% 
  group_by(Word,Category,Task) %>%
  summarize(MeanAccuracy = mean(Accuracy),
            MeanReactionTime = mean(ReactionTime))

## `summarise()` has grouped output by 'Word', 'Category'. You can override using
## the `.groups` argument.

ggplot(agr, aes(x = MeanReactionTime, y = MeanAccuracy, color = Category)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE) +
  geom_text(aes(label = Word), vjust = -0.5, hjust = 1.5) +
  facet_wrap(~Task)

## `geom_smooth()` using formula = 'y ~ x'

  # guides(legend = "none")
  # theme(legend.position = "none")  # Remove the legend

# ggplot(agr, aes(x = MeanReactionTime, y = MeanAccuracy, color = Task)) +
#   geom_point() +
#   geom_smooth(aes(color = Task), method = "lm", se = FALSE, size = 1.2) +  # Darker line
#   geom_text(aes(label = Word), vjust = -0.5, hjust = 1.5) +  
#   scale_color_manual(values = c("Adjs" = "red", 
#                                   "Nouns" = "green", 
#                                   "Verbs" = "blue")) +  # Adjust colors
#   theme(legend.position = "none")  # Remove the legend


# ggsave("../graphs/exp1b_accXrt.pdf",width = 5, height = 3)

# Compute highest accuracy for Concrete, keeping top 10 words per Category
concrete_accuracy <- d %>% 
  group_by(Category, Word, Task) %>%
  summarize(MeanAccuracy = mean(Accuracy),
            MeanReactionTime = mean(ReactionTime), .groups = "drop") %>% 
  filter(Task == "Concrete") %>%
  select(Category, Word, MeanAccuracy) %>%
  rename(ConcreteAccuracy = MeanAccuracy) %>%
  group_by(Category) %>% 
  slice_max(order_by = ConcreteAccuracy, n = 10)  # Get top 10 per category

agr <- d %>% 
  filter(Word %in% concrete_accuracy$Word) %>% 
  group_by(Word,Task,Category) %>%
  summarize(MeanAccuracy = mean(Accuracy),
            MeanReactionTime = mean(ReactionTime))

## `summarise()` has grouped output by 'Word', 'Task'. You can override using the
## `.groups` argument.

print(agr)

## # A tibble: 70 × 5
## # Groups:   Word, Task [66]
##    Word       Task     Category MeanAccuracy MeanReactionTime
##    <chr>      <chr>    <chr>           <dbl>            <dbl>
##  1 awful      Concrete Adjs            0.933             970.
##  2 awful      Valence  Adjs            0.983             772.
##  3 bleed      Concrete Verbs           0.967            1031.
##  4 bleed      Valence  Verbs           0.967             818.
##  5 cherish    Concrete Verbs           0.967            1128.
##  6 cherish    Valence  Verbs           1                 794.
##  7 chocolate  Concrete Nouns           1                 957 
##  8 chocolate  Valence  Nouns           0.933             720.
##  9 compassion Concrete Nouns           0.983             804.
## 10 compassion Valence  Nouns           0.983             834.
## # ℹ 60 more rows

ggplot(agr, aes(x = MeanReactionTime, y = MeanAccuracy)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE, color = "black") +
  geom_text(aes(label = Word, color = Task), vjust = -0.5, hjust = 1.5) +
  facet_wrap(~Category) +
  guides(legend = "none") +
  theme(legend.position = "none")  # Remove the legend

## `geom_smooth()` using formula = 'y ~ x'

# ggsave("../graphs/ConcAbs_accXrt.pdf",width = 5, height = 3)

agr <- d %>% 
  group_by(Word,Category,Task) %>%
  filter(Word %in% concrete_accuracy$Word) %>% 
  summarize(MeanAccuracy = mean(Accuracy),
            MeanReactionTime = mean(ReactionTime))

## `summarise()` has grouped output by 'Word', 'Category'. You can override using
## the `.groups` argument.

ggplot(agr, aes(x = MeanReactionTime, y = MeanAccuracy, color = Category)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE) +
  geom_text(aes(label = Word), vjust = -0.5, hjust = 1.5) +
  facet_wrap(~Task)

## `geom_smooth()` using formula = 'y ~ x'

# ggsave("../graphs/ConcAbs_catXaccXrt.pdf",width = 5, height = 2)

agr <- d %>% 
  filter(Word %in% concrete_accuracy$Word) %>% 
  group_by(Word,Task) %>%
  summarize(MeanAccuracy = mean(Accuracy),
            MeanReactionTime = mean(ReactionTime))

## `summarise()` has grouped output by 'Word'. You can override using the
## `.groups` argument.

print(agr)

## # A tibble: 66 × 4
## # Groups:   Word [33]
##    Word       Task     MeanAccuracy MeanReactionTime
##    <chr>      <chr>           <dbl>            <dbl>
##  1 awful      Concrete        0.933             970.
##  2 awful      Valence         0.983             772.
##  3 bleed      Concrete        0.967            1031.
##  4 bleed      Valence         0.967             818.
##  5 cherish    Concrete        0.967            1128.
##  6 cherish    Valence         1                 794.
##  7 chocolate  Concrete        1                 957 
##  8 chocolate  Valence         0.933             720.
##  9 compassion Concrete        0.983             804.
## 10 compassion Valence         0.983             834.
## # ℹ 56 more rows

ggplot(agr, aes(x = MeanReactionTime, y = MeanAccuracy, color = Task)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE) +  # Separate lines for each Task
  geom_text(aes(label = Word), vjust = -0.5, hjust = 1.5) +  # Task color inherited from ggplot()
  theme(legend.position = "none")  # Remove the legend

## `geom_smooth()` using formula = 'y ~ x'

# ggplot(agr, aes(x = MeanReactionTime, y = MeanAccuracy)) +
#   geom_point() +
#   geom_smooth(method = "lm", se = FALSE, color = "black") +
#   geom_text(aes(label = Word, color = Task), vjust = -0.5, hjust = 1.5) +
#   guides(legend = "none") +
#   theme(legend.position = "none")  # Remove the legend
# ggsave("../graphs/ConcAbs_accXrt.pdf",width = 5, height = 3)

agr <- d %>% 
  filter(Word %in% concrete_accuracy$Word) %>% 
  group_by(Word,Task) %>% 
  reframe(MeanAccuracy = mean(Accuracy), 
          CILow = ci.low(Accuracy), 
          CIHigh = ci.high(Accuracy)) %>%
  mutate(YMin = MeanAccuracy - CILow, 
         YMax = MeanAccuracy + CIHigh)
# View(agr)

dodge = position_dodge(.9)
ggplot(data=agr, aes(x=Word,y=MeanAccuracy, fill = Task)) +
  geom_bar(position=dodge,stat="identity") + 
  geom_errorbar(aes(ymin=YMin,ymax=YMax),width=.25,position=position_dodge(0.9)) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

agr <- d %>% 
  filter(Word %in% concrete_accuracy$Word) %>% 
  group_by(Word,Task,Category) %>% 
  reframe(MeanReactionTime = mean(ReactionTime), 
          CILow = ci.low(ReactionTime), 
          CIHigh = ci.high(ReactionTime)) %>%
  mutate(YMin = MeanReactionTime - CILow, 
         YMax = MeanReactionTime + CIHigh)
# View(agr)

dodge = position_dodge(.9)
ggplot(data=agr, aes(x=Word,y=MeanReactionTime, fill = Category)) +
  geom_bar(position=dodge,stat="identity") + 
  geom_errorbar(aes(ymin=YMin,ymax=YMax),width=.25,position=position_dodge(0.9)) +
  facet_wrap(~Task) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

In these most accurate words, is there still our effect?

names(d)

##  [1] "...1"            "ID.true"         "Word"            "Label"          
##  [5] "ConcValCombo"    "Task"            "BlockOrder"      "Group"          
##  [9] "Response"        "Accuracy"        "EventTime"       "Value"          
## [13] "RT"              "ReactionTime"    "Key_value_F"     "Key_value_J"    
## [17] "Comments"        "LogReactionTime" "LogRT"           "TrialNumber"    
## [21] "Category"

dcen <- d %>% 
  filter(Word %in% concrete_accuracy$Word) %>% 
  mutate(Word = as.factor(Word),
         ID.true = as.factor(ID.true),
         Task = as.factor(Task),
         Category = as.factor(Category),
         cAccuracy = as.numeric(Accuracy)-mean(as.numeric(Accuracy)),
         cTask = as.numeric(Task)-mean(as.numeric(Task)))

m <- lmer(LogReactionTime ~ cAccuracy*cTask + (1+cTask|Word) + (1+cTask|ID.true), data = dcen)
summary(m)

## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: LogReactionTime ~ cAccuracy * cTask + (1 + cTask | Word) + (1 +  
##     cTask | ID.true)
##    Data: dcen
## 
## REML criterion at convergence: 3098.8
## 
## Scaled residuals: 
##      Min       1Q   Median       3Q      Max 
## -10.1132  -0.5633  -0.1484   0.3715  10.3296 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev. Corr 
##  ID.true  (Intercept) 0.040580 0.20145       
##           cTask       0.026792 0.16368  -0.16
##  Word     (Intercept) 0.001337 0.03656       
##           cTask       0.005284 0.07269  0.14 
##  Residual             0.112267 0.33506       
## Number of obs: 4200, groups:  ID.true, 60; Word, 33
## 
## Fixed effects:
##                   Estimate Std. Error         df t value Pr(>|t|)    
## (Intercept)        6.68886    0.02729   64.64643 245.125  < 2e-16 ***
## cAccuracy         -0.02618    0.02988 4123.79642  -0.876 0.380975    
## cTask             -0.10804    0.02678   66.64631  -4.034 0.000143 ***
## cAccuracy:cTask   -0.06553    0.05966 4129.22350  -1.098 0.272073    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##             (Intr) cAccrc cTask 
## cAccuracy   -0.002              
## cTask       -0.108  0.021       
## cAccrcy:cTs  0.010 -0.256 -0.004

m.s <- lmer(LogReactionTime ~ cTask + (1+cTask|Word) + (1+cTask|ID.true), data = dcen)

anova(m,m.s, test="chisq")

## refitting model(s) with ML (instead of REML)

## Data: dcen
## Models:
## m.s: LogReactionTime ~ cTask + (1 + cTask | Word) + (1 + cTask | ID.true)
## m: LogReactionTime ~ cAccuracy * cTask + (1 + cTask | Word) + (1 + cTask | ID.true)
##     npar    AIC    BIC  logLik deviance  Chisq Df Pr(>Chisq)
## m.s    9 3099.6 3156.7 -1540.8   3081.6                     
## m     11 3101.0 3170.8 -1539.5   3079.0 2.6386  2     0.2673

m.c <- lmer(LogReactionTime ~ cAccuracy*cTask*Category + (1+cTask|Word) + (1+cTask|ID.true), data = dcen)
summary(m.c)

## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: LogReactionTime ~ cAccuracy * cTask * Category + (1 + cTask |  
##     Word) + (1 + cTask | ID.true)
##    Data: dcen
## 
## REML criterion at convergence: 3114.3
## 
## Scaled residuals: 
##      Min       1Q   Median       3Q      Max 
## -10.1223  -0.5641  -0.1454   0.3774  10.3234 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev. Corr 
##  ID.true  (Intercept) 0.039892 0.19973       
##           cTask       0.027225 0.16500  -0.14
##  Word     (Intercept) 0.001392 0.03731       
##           cTask       0.005248 0.07244  0.17 
##  Residual             0.112175 0.33493       
## Number of obs: 4200, groups:  ID.true, 60; Word, 33
## 
## Fixed effects:
##                                 Estimate Std. Error         df t value Pr(>|t|)
## (Intercept)                      6.66746    0.04720   64.03574 141.263  < 2e-16
## cAccuracy                       -0.10854    0.04715 4135.09816  -2.302  0.02139
## cTask                           -0.10275    0.04756   65.81477  -2.160  0.03438
## CategoryNouns                   -0.02131    0.06628   62.71939  -0.322  0.74884
## CategoryVerbs                    0.08339    0.06636   62.96993   1.257  0.21352
## cAccuracy:cTask                 -0.01895    0.09406 4123.27195  -0.202  0.84031
## cAccuracy:CategoryNouns          0.08942    0.08139 4105.92872   1.099  0.27195
## cAccuracy:CategoryVerbs          0.19142    0.07165 4114.01395   2.672  0.00758
## cTask:CategoryNouns              0.02179    0.06545   66.63913   0.333  0.74023
## cTask:CategoryVerbs             -0.03731    0.06576   66.99367  -0.567  0.57237
## cAccuracy:cTask:CategoryNouns   -0.07148    0.16256 4115.17490  -0.440  0.66019
## cAccuracy:cTask:CategoryVerbs   -0.06294    0.14303 4120.49712  -0.440  0.65993
##                                  
## (Intercept)                   ***
## cAccuracy                     *  
## cTask                         *  
## CategoryNouns                    
## CategoryVerbs                    
## cAccuracy:cTask                  
## cAccuracy:CategoryNouns          
## cAccuracy:CategoryVerbs       ** 
## cTask:CategoryNouns              
## cTask:CategoryVerbs              
## cAccuracy:cTask:CategoryNouns    
## cAccuracy:cTask:CategoryVerbs    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##             (Intr) cAccrc cTask  CtgryN CtgryV cAcc:T cAc:CN cAc:CV cTs:CN
## cAccuracy    0.014                                                        
## cTask       -0.084  0.005                                                 
## CategoryNns -0.712 -0.010  0.060                                          
## CategryVrbs -0.711 -0.010  0.060  0.512                                   
## cAccrcy:cTs  0.003 -0.107  0.028 -0.002 -0.002                            
## cAccrcy:CtN -0.008 -0.579 -0.003 -0.009  0.004  0.062                     
## cAccrcy:CtV -0.009 -0.658 -0.003  0.007  0.002  0.070  0.381              
## cTsk:CtgryN  0.061 -0.004 -0.727 -0.092 -0.041 -0.020  0.041  0.006       
## cTsk:CtgryV  0.061 -0.004 -0.723 -0.041 -0.090 -0.020  0.006  0.001  0.546
## cAccrc:T:CN -0.002  0.062 -0.016  0.020  0.003 -0.579 -0.473 -0.041 -0.018
## cAccrc:T:CV -0.002  0.070 -0.018  0.003  0.000 -0.658 -0.041 -0.038  0.015
##             cTs:CV cA:T:CN
## cAccuracy                 
## cTask                     
## CategoryNns               
## CategryVrbs               
## cAccrcy:cTs               
## cAccrcy:CtN               
## cAccrcy:CtV               
## cTsk:CtgryN               
## cTsk:CtgryV               
## cAccrc:T:CN  0.008        
## cAccrc:T:CV  0.003  0.381

Look within categorieslexical

Adjectives

d$Category = as.factor(d$Category)
str(d$Category)

##  Factor w/ 3 levels "Adjs","Nouns",..: 3 3 3 3 3 3 3 3 3 3 ...

dcen <- d %>% 
  filter(Word %in% concrete_accuracy$Word) %>% 
  filter(Category == "Adjs") %>% 
  mutate(Word = as.factor(Word),
         ID.true = as.factor(ID.true),
         Task = as.factor(Task),
         Category = as.factor(Category),
         cAccuracy = as.numeric(Accuracy)-mean(as.numeric(Accuracy)),
         cTask = as.numeric(Task)-mean(as.numeric(Task)))

m <- lmer(LogReactionTime ~ cAccuracy*cTask + (1+cTask|Word) + (1+cTask|ID.true), data = dcen)

## boundary (singular) fit: see help('isSingular')

## Warning: Model failed to converge with 1 negative eigenvalue: -7.8e+00

summary(m)

## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: LogReactionTime ~ cAccuracy * cTask + (1 + cTask | Word) + (1 +  
##     cTask | ID.true)
##    Data: dcen
## 
## REML criterion at convergence: 856.9
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -3.1537 -0.5446 -0.1372  0.3511  7.5771 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev. Corr 
##  ID.true  (Intercept) 0.048745 0.22078       
##           cTask       0.040973 0.20242  -0.34
##  Word     (Intercept) 0.000000 0.00000       
##           cTask       0.004243 0.06514   NaN 
##  Residual             0.108071 0.32874       
## Number of obs: 1200, groups:  ID.true, 20; Word, 10
## 
## Fixed effects:
##                   Estimate Std. Error         df t value Pr(>|t|)    
## (Intercept)        6.66903    0.05027   18.99442 132.657   <2e-16 ***
## cAccuracy         -0.11017    0.04628 1178.29554  -2.381   0.0174 *  
## cTask             -0.10249    0.05323   21.29784  -1.925   0.0676 .  
## cAccuracy:cTask   -0.01873    0.09207 1175.88757  -0.203   0.8388    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##             (Intr) cAccrc cTask 
## cAccuracy    0.000              
## cTask       -0.285  0.007       
## cAccrcy:cTs  0.004 -0.114 -0.001
## optimizer (nloptwrap) convergence code: 0 (OK)
## boundary (singular) fit: see help('isSingular')

m.s <- lmer(LogReactionTime ~ cTask + (1+cTask|Word) + (1+cTask|ID.true), data = dcen)

## boundary (singular) fit: see help('isSingular')

## Warning: Model failed to converge with 1 negative eigenvalue: -1.8e+00

anova(m,m.s, test="chisq")

## refitting model(s) with ML (instead of REML)

## Data: dcen
## Models:
## m.s: LogReactionTime ~ cTask + (1 + cTask | Word) + (1 + cTask | ID.true)
## m: LogReactionTime ~ cAccuracy * cTask + (1 + cTask | Word) + (1 + cTask | ID.true)
##     npar   AIC    BIC  logLik deviance  Chisq Df Pr(>Chisq)  
## m.s    9 865.2 911.01 -423.60    847.2                       
## m     11 863.3 919.29 -420.65    841.3 5.9034  2    0.05225 .
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Verbs

dcen <- d %>% 
  filter(Word %in% concrete_accuracy$Word) %>% 
  filter(Category == "Verbs") %>% 
  mutate(Word = as.factor(Word),
         ID.true = as.factor(ID.true),
         Task = as.factor(Task),
         Category = as.factor(Category),
         cAccuracy = as.numeric(Accuracy)-mean(as.numeric(Accuracy)),
         cTask = as.numeric(Task)-mean(as.numeric(Task)))

m <- lmer(LogReactionTime ~ cAccuracy*cTask + (1+cTask|Word) + (1+cTask|ID.true), data = dcen)
summary(m)

## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: LogReactionTime ~ cAccuracy * cTask + (1 + cTask | Word) + (1 +  
##     cTask | ID.true)
##    Data: dcen
## 
## REML criterion at convergence: 1056.9
## 
## Scaled residuals: 
##      Min       1Q   Median       3Q      Max 
## -10.0886  -0.5691  -0.1218   0.4304  10.3293 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev. Corr 
##  ID.true  (Intercept) 0.043491 0.20855       
##           cTask       0.025435 0.15948  0.12 
##  Word     (Intercept) 0.003336 0.05775       
##           cTask       0.005753 0.07585  -0.39
##  Residual             0.110844 0.33293       
## Number of obs: 1440, groups:  ID.true, 20; Word, 12
## 
## Fixed effects:
##                   Estimate Std. Error         df t value Pr(>|t|)    
## (Intercept)        6.75085    0.05029   22.98100 134.227  < 2e-16 ***
## cAccuracy          0.08468    0.05364 1399.56925   1.579  0.11462    
## cTask             -0.15476    0.04538   22.56421  -3.410  0.00244 ** 
## cAccuracy:cTask   -0.08913    0.10714 1404.07656  -0.832  0.40563    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##             (Intr) cAccrc cTask
## cAccuracy   0.000              
## cTask       0.027  0.000       
## cAccrcy:cTs 0.000  0.014  0.000

m.s <- lmer(LogReactionTime ~ cTask + (1+cTask|Word) + (1+cTask|ID.true), data = dcen)

anova(m,m.s, test="chisq")

## refitting model(s) with ML (instead of REML)

## Data: dcen
## Models:
## m.s: LogReactionTime ~ cTask + (1 + cTask | Word) + (1 + cTask | ID.true)
## m: LogReactionTime ~ cAccuracy * cTask + (1 + cTask | Word) + (1 + cTask | ID.true)
##     npar    AIC    BIC  logLik deviance  Chisq Df Pr(>Chisq)
## m.s    9 1062.9 1110.4 -522.46   1044.9                     
## m     11 1063.7 1121.7 -520.86   1041.7 3.2031  2     0.2016

Nouns

dcen <- d %>% 
  filter(Word %in% concrete_accuracy$Word) %>% 
  filter(Category == "Nouns") %>% 
  mutate(Word = as.factor(Word),
         ID.true = as.factor(ID.true),
         Task = as.factor(Task),
         Category = as.factor(Category),
         cAccuracy = as.numeric(Accuracy)-mean(as.numeric(Accuracy)),
         cTask = as.numeric(Task)-mean(as.numeric(Task)))

m <- lmer(LogReactionTime ~ cAccuracy*cTask + (1+cTask|Word) + (1+cTask|ID.true), data = dcen)

## boundary (singular) fit: see help('isSingular')

summary(m)

## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: LogReactionTime ~ cAccuracy * cTask + (1 + cTask | Word) + (1 +  
##     cTask | ID.true)
##    Data: dcen
## 
## REML criterion at convergence: 1180.1
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -2.4028 -0.5825 -0.1672  0.3682  9.4460 
## 
## Random effects:
##  Groups   Name        Variance  Std.Dev. Corr 
##  ID.true  (Intercept) 0.0276381 0.16625       
##           cTask       0.0165818 0.12877  -0.16
##  Word     (Intercept) 0.0009145 0.03024       
##           cTask       0.0057894 0.07609  1.00 
##  Residual             0.1160707 0.34069       
## Number of obs: 1560, groups:  ID.true, 20; Word, 13
## 
## Fixed effects:
##                   Estimate Std. Error         df t value Pr(>|t|)    
## (Intercept)        6.64526    0.03910   20.64477 169.956   <2e-16 ***
## cAccuracy         -0.02485    0.06721 1520.21240  -0.370   0.7116    
## cTask             -0.07492    0.03976   22.47093  -1.884   0.0725 .  
## cAccuracy:cTask   -0.08226    0.13434 1523.63135  -0.612   0.5404    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##             (Intr) cAccrc cTask 
## cAccuracy   -0.025              
## cTask        0.004  0.074       
## cAccrcy:cTs  0.037 -0.655 -0.048
## optimizer (nloptwrap) convergence code: 0 (OK)
## boundary (singular) fit: see help('isSingular')

m.s <- lmer(LogReactionTime ~ cTask + (1+cTask|Word) + (1+cTask|ID.true), data = dcen)

## boundary (singular) fit: see help('isSingular')

## Warning: Model failed to converge with 1 negative eigenvalue: -3.9e+01

anova(m,m.s, test="chisq")

## refitting model(s) with ML (instead of REML)

## Data: dcen
## Models:
## m.s: LogReactionTime ~ cTask + (1 + cTask | Word) + (1 + cTask | ID.true)
## m: LogReactionTime ~ cAccuracy * cTask + (1 + cTask | Word) + (1 + cTask | ID.true)
##     npar    AIC    BIC  logLik deviance  Chisq Df Pr(>Chisq)
## m.s    9 1183.9 1232.1 -582.95   1165.9                     
## m     11 1186.5 1245.4 -582.24   1164.5 1.4231  2     0.4909

By Participant

agr <- d %>%
    group_by(ID.true,Task) %>%
    reframe(MeanAccuracy = mean(Accuracy), 
            CILow = ci.low(Accuracy), 
            CIHigh = ci.high(Accuracy)) %>%
    mutate(YMin = MeanAccuracy - CILow, 
           YMax = MeanAccuracy + CIHigh)

dodge = position_dodge(.9)
ggplot(data=agr, aes(x=reorder(ID.true,MeanAccuracy),y=MeanAccuracy,fill=Task)) +
  geom_bar(position=dodge,stat="identity") +
  facet_wrap(~Task) +
  geom_errorbar(aes(ymin=YMin,ymax=YMax),width=.25,position=position_dodge(0.9)) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

  # guides(fill = "none")

ReactionTime by Task

agr <- d %>% 
  group_by(Word,Task,Category) %>% 
  reframe(MeanAccuracy = mean(Accuracy), 
          CILow = ci.low(Accuracy), 
          CIHigh = ci.high(Accuracy)) %>%
  mutate(YMin = MeanAccuracy - CILow, 
         YMax = MeanAccuracy + CIHigh)

ggplot(agr, aes(x=Task, y=MeanAccuracy,fill=Category)) + 
    geom_violin(trim=FALSE,alpha=.4) +
    geom_jitter(shape=16, position=position_jitter(0.2))

  # guides(fill = "none")

By Participant

agr <- d %>%
    group_by(ID.true,Task,ConcValCombo) %>%
    summarize(MeanReactionTime = mean(ReactionTime), 
              CILow = ci.low(ReactionTime), 
              CIHigh = ci.high(ReactionTime)) %>%
    mutate(YMin = MeanReactionTime - CILow, 
           YMax = MeanReactionTime + CIHigh)

## `summarise()` has grouped output by 'ID.true', 'Task'. You can override using
## the `.groups` argument.

dodge = position_dodge(.9)
ggplot(data=agr, aes(x=reorder(ID.true,MeanReactionTime),y=MeanReactionTime,fill=ConcValCombo)) +
  geom_bar(position=dodge,stat="identity") +
  facet_wrap(~Task) +
  geom_errorbar(aes(ymin=YMin,ymax=YMax),width=.25,position=position_dodge(0.9)) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

  # guides(fill = "none")

Looking at Responses Generally

Proportion Concrete

agr <- d %>% 
  filter(Task == "Concrete") %>% 
  mutate(Response.n = as.numeric(factor(Response, levels = c("abstract", "concrete"))) - 1) %>% 
  group_by(ConcValCombo,Category) %>%
  summarize(PropConcrete = mean(Response.n),
          CILow = ci.low(Response.n), 
          CIHigh = ci.high(Response.n)) %>%
  mutate(YMin = PropConcrete - CILow, 
          YMax = PropConcrete + CIHigh)

## `summarise()` has grouped output by 'ConcValCombo'. You can override using the
## `.groups` argument.

dodge = position_dodge(.9)
ggplot(data=agr, aes(x=Category,y=PropConcrete,fill=ConcValCombo)) +
  geom_bar(position=dodge,stat="identity") +
  # facet_wrap(~Version) +
  geom_errorbar(aes(ymin=YMin,ymax=YMax),width=.25,position=position_dodge(0.9))

  # theme(axis.text.x = element_text(angle = 45, hjust = 1))

agr <- d %>% 
  filter(Task == "Concrete") %>% 
  mutate(Response.n = as.numeric(factor(Response, levels = c("abstract", "concrete"))) - 1) %>% 
  group_by(Word,ConcValCombo) %>%
  summarize(PropConcrete = mean(Response.n),
            MeanReactionTime = mean(ReactionTime))

## `summarise()` has grouped output by 'Word'. You can override using the
## `.groups` argument.

ggplot(agr, aes(x = MeanReactionTime, y = PropConcrete)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE, color = "red") +
  geom_text(aes(label = Word, color = ConcValCombo), vjust = -0.5, hjust = 1.5)

## `geom_smooth()` using formula = 'y ~ x'

  # geom_text_repel(aes(label = Word, color = ConcValCombo), 
                  # vjust = -0.5, hjust = 1.5) +
  # scale_fill_manual(values=cbPalette) +
  # scale_color_manual(values=cbPalette)

agr <- d %>% 
  filter(Task == "Valence") %>% 
  mutate(Response.n = as.numeric(factor(Response, levels = c("negative", "positive"))) - 1) %>% 
  group_by(Word,ConcValCombo) %>%
  summarize(PropPositive = mean(Response.n),
            MeanReactionTime = mean(ReactionTime))

## `summarise()` has grouped output by 'Word'. You can override using the
## `.groups` argument.

ggplot(agr, aes(x = MeanReactionTime, y = PropPositive)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE, color = "red") +
  geom_text(aes(label = Word, color = ConcValCombo), vjust = -0.5, hjust = 1.5)

## `geom_smooth()` using formula = 'y ~ x'

  # geom_text_repel(aes(label = Word, color = ConcValCombo), 
                  # vjust = -0.5, hjust = 1.5) +
  # scale_fill_manual(values=cbPalette) +
  # scale_color_manual(values=cbPalette)

agr <- d %>% 
  # filter(Task == "Concrete") %>% 
  # mutate(Response.n = as.numeric(factor(Response, levels = c("abstract", "concrete"))) - 1) %>% 
  group_by(Word,Task,Category) %>%
  summarize(MeanAccuracy = mean(Accuracy),
            MeanReactionTime = mean(ReactionTime))

## `summarise()` has grouped output by 'Word', 'Task'. You can override using the
## `.groups` argument.

ggplot(agr, aes(x = MeanReactionTime, y = MeanAccuracy)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE, color = "black") +
  facet_wrap(~Task) +
  geom_text(aes(label = Word, color = Category), vjust = -0.5, hjust = 1.5) +
  guides(legend = "none")

## `geom_smooth()` using formula = 'y ~ x'

  #   theme(
  #   legend.position = "top",    # Move legend to the top
  #   legend.title = element_text(size = 10),  # Adjust legend title size
  #   legend.text = element_text(size = 9)     # Adjust legend text size
  # )
  

  # geom_text_repel(aes(label = Word, color = ConcValCombo), 
                  # vjust = -0.5, hjust = 1.5) +
  # scale_fill_manual(values=cbPalette) +
  # scale_color_manual(values=cbPalette)

# ggsave("../graphs/exp3_accXrt.pdf",width = 5, height = 3)

Reaction Time analyses

Data Cleaning

# First remove inaccurate participants
length(unique(d$ID.true))

## [1] 60

inacc.parts <- d %>% 
  group_by(ID.true,Task,Category) %>% 
  summarise(MeanAccuracy = mean(Accuracy)) %>% 
  filter(MeanAccuracy < .75)

## `summarise()` has grouped output by 'ID.true', 'Task'. You can override using
## the `.groups` argument.

# How many participants have Accuracy < .75?
length(unique(inacc.parts$ID.true))

## [1] 8

# Remove them
d.inaccurate.removed <- d %>% 
  anti_join(inacc.parts, by = "ID.true")
# Sanity check
length(unique(d.inaccurate.removed$ID.true))

## [1] 52

# Second, remove all inaccurate trials
orig <- nrow(d.inaccurate.removed)
d.inaccurate.removed <- d.inaccurate.removed %>% 
  filter(Accuracy == 1)
nrow(d.inaccurate.removed)/orig*100

## [1] 93.46154

# Third, Remove subjects with ReactionTime higher than 3x IQR
summary(d.inaccurate.removed$LogReactionTime)

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##   2.303   6.435   6.613   6.705   6.886  10.619

  #  Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
  # 6.924   7.328   7.436   7.479   7.579  10.008 
range(d.inaccurate.removed$LogReactionTime)

## [1]  2.302585 10.618714

hist(d.inaccurate.removed$LogReactionTime, breaks=100, col="lightblue", xlab="LogReactionTime (ms)",
        main="Histogram with Normal Curve")

quantile(d.inaccurate.removed$LogReactionTime, na.rm = TRUE)

##        0%       25%       50%       75%      100% 
##  2.302585  6.434547  6.613384  6.885765 10.618714

IQR(d.inaccurate.removed$LogReactionTime, na.rm = TRUE)*3 # 0.7526289

## [1] 1.353656

cutoff.high <- quantile(d.inaccurate.removed$LogReactionTime, na.rm = TRUE)[4] + IQR(d.inaccurate.removed$LogReactionTime, na.rm = TRUE)*3 # 8.419261
cutoff.low <- quantile(d.inaccurate.removed$LogReactionTime, na.rm = TRUE)[2] - IQR(d.inaccurate.removed$LogReactionTime, na.rm = TRUE)*3# 6.5088838.419261


# remove subjects with ReactionTime higher than 3 x IQR
df.outliers.removed <- subset(d.inaccurate.removed, (d.inaccurate.removed$LogReactionTime > cutoff.low) & (d.inaccurate.removed$LogReactionTime < cutoff.high))

hist(df.outliers.removed$LogReactionTime, breaks=100, col="lightblue", xlab="LogReactionTime (ms)",
        main="Histogram with Normal Curve")

ggplot(df.outliers.removed, aes(x=LogReactionTime, fill=Task)) +
  # facet_wrap(~BlockOrder) +
  geom_density(alpha = .4)

ggplot(df.outliers.removed, aes(x=ReactionTime, fill=Task)) +
  # facet_wrap(~BlockOrder) +
  geom_density(alpha = .4)

Summary Stats

agr <- d.inaccurate.removed %>% 
  group_by(Task,Category) %>% 
  summarize(MeanRT = mean(ReactionTime),
            SD = sd(ReactionTime),
            MeanLogRT = mean(LogReactionTime))

## `summarise()` has grouped output by 'Task'. You can override using the
## `.groups` argument.

print(agr)

## # A tibble: 6 × 5
## # Groups:   Task [2]
##   Task     Category MeanRT    SD MeanLogRT
##   <chr>    <fct>     <dbl> <dbl>     <dbl>
## 1 Concrete Adjs      1033.  762.      6.81
## 2 Concrete Nouns      918.  715.      6.72
## 3 Concrete Verbs     1071. 1091.      6.85
## 4 Valence  Adjs       799.  387.      6.61
## 5 Valence  Nouns      773.  687.      6.57
## 6 Valence  Verbs      866.  406.      6.69

LogReactionTime by Task

agr <- df.outliers.removed %>%
    group_by(Task,Word) %>%
    summarize(MeanLogReactionTime = mean(LogReactionTime), 
              CILow = ci.low(LogReactionTime), 
              CIHigh = ci.high(LogReactionTime)) %>%
    mutate(YMin = MeanLogReactionTime - CILow, 
           YMax = MeanLogReactionTime + CIHigh)

## `summarise()` has grouped output by 'Task'. You can override using the
## `.groups` argument.

ggplot(agr, aes(x=MeanLogReactionTime, fill=Task)) +
  geom_density(alpha = .4)

ggplot(agr, aes(x=Task, y=MeanLogReactionTime,fill=Task)) + 
    geom_violin(trim=FALSE,alpha=.4) +
    geom_jitter(shape=16, position=position_jitter(0.2)) +
    # geom_errorbar(aes(ymin=YMin,ymax=YMax),width=.25,position="dodge", show.legend = FALSE) +
  # theme(axis.text.x = element_text(angle = 45, hjust = 1)) +
  guides(fill = "none")

ReactionTime by Task

agr <- df.outliers.removed %>%
    group_by(Task,Word) %>%
    summarize(MeanReactionTime = mean(ReactionTime), 
              CILow = ci.low(ReactionTime), 
              CIHigh = ci.high(ReactionTime)) %>%
    mutate(YMin = MeanReactionTime - CILow, 
           YMax = MeanReactionTime + CIHigh)

## `summarise()` has grouped output by 'Task'. You can override using the
## `.groups` argument.

ggplot(agr, aes(x=MeanReactionTime, fill=Task)) +
  geom_density(alpha = .4)

ggplot(agr, aes(x=Task, y=MeanReactionTime,fill=Task)) + 
    geom_violin(trim=FALSE,alpha=.4) +
    geom_jitter(shape=16, position=position_jitter(0.2)) +
    # geom_errorbar(aes(ymin=YMin,ymax=YMax),width=.25,position="dodge", show.legend = FALSE) +
  # theme(axis.text.x = element_text(angle = 45, hjust = 1)) +
  guides(fill = "none")

# ggsave("../graphs/total_rt_violin.pdf",width = 4, height = 3)

Prop Conc/Val

prop <- d %>% 
  filter(Task == "Concrete") %>% 
  mutate(Response.n = as.numeric(factor(Response, levels = c("abstract", "concrete"))) - 1) %>% 
  group_by(Word) %>%
  summarize(PropConcrete = mean(Response.n))


agr <- df.outliers.removed %>% 
  filter(Task == "Concrete") %>% 
  mutate(Response.n = as.numeric(factor(Response, levels = c("abstract", "concrete"))) - 1) %>% 
  group_by(Word,ConcValCombo) %>%
  summarize(PropConcrete = mean(Response.n),
            MeanReactionTime = mean(ReactionTime))

## `summarise()` has grouped output by 'Word'. You can override using the
## `.groups` argument.

ggplot(agr, aes(x = MeanReactionTime, y = PropConcrete)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE, color = "red") +
  geom_text(aes(label = Word,color=ConcValCombo), vjust = -0.5, hjust = 1.5)

## `geom_smooth()` using formula = 'y ~ x'

  # geom_text_repel(aes(label = Word, color = ConcValCombo), 
                  # vjust = -0.5, hjust = 1.5) +
  # scale_fill_manual(values=cbPalette) +
  # scale_color_manual(values=cbPalette)

Does REsponse choice influence RT??

Valence

Yes, looks like choosing negative is faster than choosing positive d

Concrete

just barely, choosing abstract has a negative effect on RT

ReactionTime by Task

agr <- df.outliers.removed %>%
    group_by(Word,Task,Category) %>%
    reframe(MeanReactionTime = mean(ReactionTime), 
            CILow = ci.low(ReactionTime), 
            CIHigh = ci.high(ReactionTime)) %>%
    mutate(YMin = MeanReactionTime - CILow, 
           YMax = MeanReactionTime + CIHigh)

ggplot(agr, aes(x=Category, y=MeanReactionTime,fill=Task)) + 
    geom_violin(trim=FALSE,alpha=.4) +
    geom_jitter(shape=16, position=position_jitter(0.2)) +
    guides(fill = "none")

Look at cases where RT Conc <= RT Valence or Conc Acc >= Val Acc

test <- df.outliers.removed %>%
  # filter(Task %in% c("Concrete", "Valence")) %>%  # Keep only relevant tasks
  group_by(Word, Task) %>%
  summarise(
    RT = mean(ReactionTime, na.rm = TRUE),  # Take the mean RT if duplicates exist
    .groups = "drop_last" # Drop grouping by Task but keep Word and ID.true
  ) %>%
  pivot_wider(names_from = Task, values_from = RT, names_prefix = "RT_") %>% # Reshape to wide format
  filter(RT_Concrete <= RT_Valence) %>%  # Apply the condition
    pivot_longer(
    cols = starts_with("RT_"), # Select the reshaped columns
    names_to = "Task",         # Restore Task column
    names_prefix = "RT_",      # Remove "RT_" prefix to match original Task names
    values_to = "ReactionTime" # Column for the RT values
  ) %>%
  ungroup()

nrow(test)/nrow(df.outliers.removed)

## [1] 0.001553733

agr <- test %>% 
    group_by(Word, Task) %>%
    summarize(MeanReactionTime = mean(ReactionTime), 
            CILow = ci.low(ReactionTime), 
            CIHigh = ci.high(ReactionTime)) %>%
    mutate(YMin = MeanReactionTime - CILow, 
           YMax = MeanReactionTime + CIHigh)

## `summarise()` has grouped output by 'Word'. You can override using the
## `.groups` argument.

dodge = position_dodge(.9)
ggplot(data=agr, aes(x=Task,y=MeanReactionTime,fill=Task)) +
  geom_bar(position=dodge,stat="identity") +
  facet_wrap(~Word) +
  geom_errorbar(aes(ymin=YMin,ymax=YMax),width=.25,position=position_dodge(0.9)) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

  # guides(fill = "none")

ggplot(agr,aes(x=Word, y=MeanReactionTime, alpha=Task, fill=Task)) +
  geom_bar(position="dodge",stat="identity") +
  geom_errorbar(aes(ymin=YMin,ymax=YMax),width=.25,position=position_dodge(0.9)) +
  # facet_wrap(~Task, ncol=10) +
  xlab("ConcValCombo") +
  ylab("MeanAccuracy") +
  guides(fill=FALSE) +
  # guides(alpha=guide_legend(title="Task")) +
  theme(legend.key.size = unit(0.3, "cm"),
        legend.position = "top", # c(.5,1)
        legend.direction = "horizontal",
        legend.margin=margin(0,0,0,0),
        legend.box.margin=margin(0,0,-5,-5),legend.spacing.y = unit(0.001, 'cm')) +
    # scale_fill_manual(values=cbPalette) +
    # scale_color_manual(values=cbPalette) +
    scale_alpha_discrete(range = c(.5,1)) +
    theme(axis.text.x = element_text(angle = 45, hjust = 1))

## Warning: The `<scale>` argument of `guides()` cannot be `FALSE`. Use "none" instead as
## of ggplot2 3.3.4.
## This warning is displayed once every 8 hours.
## Call `lifecycle::last_lifecycle_warnings()` to see where this warning was
## generated.

## Warning: Using alpha for a discrete variable is not advised.

test_avv <- d %>% 
  filter(Word %in% test$Word) %>% 
  group_by(Word, Task) %>%
    summarize(MeanAccuracy = mean(Accuracy),
            CILow = ci.low(Accuracy), 
            CIHigh = ci.high(Accuracy),
            ) %>%
    mutate(YMin = MeanAccuracy - CILow, 
           YMax = MeanAccuracy + CIHigh)

## `summarise()` has grouped output by 'Word'. You can override using the
## `.groups` argument.

# View(test_avv)

dodge = position_dodge(.9)
ggplot(data=test_avv, aes(x=Task,y=MeanAccuracy,fill=Task)) +
  geom_bar(position=dodge,stat="identity") +
  facet_wrap(~Word) +
  geom_errorbar(aes(ymin=YMin,ymax=YMax),width=.25,position=position_dodge(0.9)) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

  # guides(fill = "none")

test_avv <- d %>% 
  filter(Word %in% test$Word) %>% 
  group_by(Word, Task) %>%
    summarize(MeanAccuracy = mean(Accuracy),
              MeanReactionTime = mean(ReactionTime))

## `summarise()` has grouped output by 'Word'. You can override using the
## `.groups` argument.

ggplot(test_avv, aes(x = MeanReactionTime, y = MeanAccuracy)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE, color = "red") +
  # geom_text(aes(label = Word, color = Task), vjust = -0.5, hjust = 1.5)
  geom_text_repel(aes(label = Word, color = Task), 
                  vjust = -0.5, hjust = 1.5) +
  scale_fill_manual(values=cbPalette) +
  scale_color_manual(values=cbPalette)

## `geom_smooth()` using formula = 'y ~ x'

Conc Acc is higher than Val Acc

# Compute highest accuracy for Concrete
concrete_accuracy <- d %>% 
  group_by(Word,Task) %>%
  summarize(MeanAccuracy = mean(Accuracy),
            MeanReactionTime = mean(ReactionTime)) %>% 
  filter(Task == "Concrete") %>%
  select(Word, MeanAccuracy) %>%
  rename(ConcreteAccuracy = MeanAccuracy) %>% 
  arrange(desc(ConcreteAccuracy)) %>% 
  head(10)

## `summarise()` has grouped output by 'Word'. You can override using the
## `.groups` argument.

agr <- df.outliers.removed %>% 
  filter(Word %in% concrete_accuracy$Word) %>% 
  group_by(Word,Task) %>%
  summarize(MeanAccuracy = mean(Accuracy),
            MeanReactionTime = mean(ReactionTime))

## `summarise()` has grouped output by 'Word'. You can override using the
## `.groups` argument.

print(agr)

## # A tibble: 20 × 4
## # Groups:   Word [10]
##    Word        Task     MeanAccuracy MeanReactionTime
##    <chr>       <chr>           <dbl>            <dbl>
##  1 chocolate   Concrete            1             747.
##  2 chocolate   Valence             1             697.
##  3 compassion  Concrete            1             788.
##  4 compassion  Valence             1             815.
##  5 despair     Concrete            1             818.
##  6 despair     Valence             1             746.
##  7 fireplace   Concrete            1             771.
##  8 fireplace   Valence             1             813.
##  9 grief       Concrete            1             861.
## 10 grief       Valence             1             769.
## 11 inspiration Concrete            1             800.
## 12 inspiration Valence             1             782.
## 13 kindness    Concrete            1             947.
## 14 kindness    Valence             1             635.
## 15 kiss        Concrete            1             877.
## 16 kiss        Valence             1             831.
## 17 mud         Concrete            1             906.
## 18 mud         Valence             1             770.
## 19 shame       Concrete            1             849.
## 20 shame       Valence             1             705.

ggplot(agr, aes(x = MeanReactionTime, y = MeanAccuracy)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE, color = "black") +
  geom_text(aes(label = Word, color = Task), vjust = -0.5, hjust = 1.5)

## `geom_smooth()` using formula = 'y ~ x'

  # guides(legend = "none")
  # theme(legend.position = "none")  # Remove the legend
# ggsave("../graphs/exp1b_accXrt.pdf",width = 5, height = 3)

agr <- df.outliers.removed %>% 
    filter(Word %in% concrete_accuracy$Word) %>% 
    group_by(Word, Task) %>%
    summarize(MeanReactionTime = mean(ReactionTime), 
            CILow = ci.low(ReactionTime), 
            CIHigh = ci.high(ReactionTime)) %>%
    mutate(YMin = MeanReactionTime - CILow, 
           YMax = MeanReactionTime + CIHigh)

## `summarise()` has grouped output by 'Word'. You can override using the
## `.groups` argument.

print(agr)

## # A tibble: 20 × 7
## # Groups:   Word [10]
##    Word        Task     MeanReactionTime CILow CIHigh  YMin  YMax
##    <chr>       <chr>               <dbl> <dbl>  <dbl> <dbl> <dbl>
##  1 chocolate   Concrete             747.  67.1   75.9  680.  823.
##  2 chocolate   Valence              697.  52.6   54.4  645.  752.
##  3 compassion  Concrete             788.  60.4   73.4  728.  861.
##  4 compassion  Valence              815. 101.   122.   714.  936.
##  5 despair     Concrete             818. 101.   140.   717.  958.
##  6 despair     Valence              746.  61.5   70.6  684.  816.
##  7 fireplace   Concrete             771.  60.9   70.5  710.  842.
##  8 fireplace   Valence              813.  66.7   73.0  746.  886.
##  9 grief       Concrete             861.  96.8  114.   764.  976.
## 10 grief       Valence              769.  87.6  110.   682.  880.
## 11 inspiration Concrete             800.  67.4   73.3  733.  874.
## 12 inspiration Valence              782.  63.3   73.8  718.  855.
## 13 kindness    Concrete             947. 132.   146.   816. 1094.
## 14 kindness    Valence              635.  35.4   34.5  599.  669.
## 15 kiss        Concrete             877.  97.9  112.   779.  990.
## 16 kiss        Valence              831.  92.2  114.   739.  945.
## 17 mud         Concrete             906. 124.   126.   782. 1033.
## 18 mud         Valence              770.  80.8   94.7  689.  864.
## 19 shame       Concrete             849.  84.5   94.0  764.  943.
## 20 shame       Valence              705.  45.3   48.5  660.  754.

dodge = position_dodge(.9)
ggplot(data=agr, aes(x=Word,y=MeanReactionTime,fill=Task)) +
  geom_bar(position=dodge,stat="identity") +
  # facet_wrap(~Word) +
  geom_errorbar(aes(ymin=YMin,ymax=YMax),width=.25,position=position_dodge(0.9)) +
  theme(axis.text.x = element_text(angle = 45, hjust = 1))

  # guides(fill = "none")

agr <- df.outliers.removed %>%
  filter(Word %in% concrete_accuracy$Word)

ggplot(agr, aes(x=Word, y=ReactionTime,fill=Task)) + 
    geom_violin(trim=FALSE,alpha=.4) +
    # Median dot
    stat_summary(fun = median, geom = "point", 
                 shape = 21, size = 1.5, 
                 position = position_dodge(width=0.9)) +  # Centering the median dot
    # geom_jitter(shape=10, position=position_jitter(0.2)) +
    theme(axis.text.x = element_text(angle = 45, hjust = 1))

    # guides(fill = "none")

Analysis

convert everything to factors

Is there a difference between Semantic and Valence Tasks?

Yes

m = lmer(LogReactionTime ~ Task + (1|ID.true) + (1|Word), data=center)
summary(m)

## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: LogReactionTime ~ Task + (1 | ID.true) + (1 | Word)
##    Data: center
## 
## REML criterion at convergence: 5191.1
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -3.5057 -0.6543 -0.1714  0.4628  5.0814 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev.
##  Word     (Intercept) 0.002588 0.05087 
##  ID.true  (Intercept) 0.036627 0.19138 
##  Residual             0.088528 0.29754 
## Number of obs: 11585, groups:  Word, 117; ID.true, 52
## 
## Fixed effects:
##               Estimate Std. Error         df t value Pr(>|t|)    
## (Intercept)  6.776e+00  2.726e-02  5.530e+01  248.60   <2e-16 ***
## TaskValence -1.555e-01  5.546e-03  1.145e+04  -28.04   <2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##             (Intr)
## TaskValence -0.105

table(center$Category)

## 
##  Adjs Nouns Verbs 
##  3073  4239  4273

m = lmer(LogReactionTime ~ Category + (1|ID.true) + (1|Word), data=center)
summary(m)

## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: LogReactionTime ~ Category + (1 | ID.true) + (1 | Word)
##    Data: center
## 
## REML criterion at convergence: 5946.5
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -3.1338 -0.6525 -0.1809  0.4647  5.1642 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev.
##  Word     (Intercept) 0.002364 0.04862 
##  ID.true  (Intercept) 0.035268 0.18780 
##  Residual             0.094659 0.30767 
## Number of obs: 11585, groups:  Word, 117; ID.true, 52
## 
## Fixed effects:
##               Estimate Std. Error       df t value Pr(>|t|)    
## (Intercept)    6.70018    0.05108 51.28017 131.170   <2e-16 ***
## CategoryNouns -0.06628    0.06739 51.49656  -0.984    0.330    
## CategoryVerbs  0.05418    0.06739 51.49045   0.804    0.425    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##             (Intr) CtgryN
## CategoryNns -0.758       
## CategryVrbs -0.758  0.576

m = lmer(LogReactionTime ~ Task*Category + (1|ID.true) + (1|Word), data=center)
summary(m)

## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: LogReactionTime ~ Task * Category + (1 | ID.true) + (1 | Word)
##    Data: center
## 
## REML criterion at convergence: 5202.6
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -3.4944 -0.6547 -0.1680  0.4643  5.0992 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev.
##  Word     (Intercept) 0.002581 0.0508  
##  ID.true  (Intercept) 0.035509 0.1884  
##  Residual             0.088497 0.2975  
## Number of obs: 11585, groups:  Word, 117; ID.true, 52
## 
## Fixed effects:
##                             Estimate Std. Error         df t value Pr(>|t|)    
## (Intercept)                6.793e+00  5.159e-02  5.273e+01 131.679   <2e-16 ***
## TaskValence               -1.781e-01  1.078e-02  1.145e+04 -16.523   <2e-16 ***
## CategoryNouns             -8.143e-02  6.806e-02  5.293e+01  -1.197   0.2368    
## CategoryVerbs              3.424e-02  6.805e-02  5.290e+01   0.503   0.6169    
## TaskValence:CategoryNouns  2.804e-02  1.416e-02  1.146e+04   1.980   0.0477 *  
## TaskValence:CategoryVerbs  3.340e-02  1.411e-02  1.144e+04   2.366   0.0180 *  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##             (Intr) TskVln CtgryN CtgryV TsV:CN
## TaskValence -0.109                            
## CategoryNns -0.758  0.083                     
## CategryVrbs -0.758  0.083  0.577              
## TskVlnc:CtN  0.083 -0.761 -0.108 -0.063       
## TskVlnc:CtV  0.083 -0.764 -0.063 -0.107  0.582

In the Concreteness task, is there a difference between concreteness and abstractness on ReactionTime?

Nope

str(df_factors)

## tibble [11,585 × 21] (S3: tbl_df/tbl/data.frame)
##  $ ...1           : Factor w/ 4775 levels "1","2","3","4",..: 1 3 4 5 6 7 8 9 10 11 ...
##  $ ID.true        : Factor w/ 52 levels "5588b04ffdf99b7a91e75ddb",..: 5 5 5 5 5 5 5 5 5 5 ...
##  $ Word           : Factor w/ 117 levels "admired","annoy",..: 63 86 2 12 90 6 70 101 57 114 ...
##  $ Label          : Factor w/ 2 levels "test_conc","test_val": 2 2 2 2 2 2 2 2 2 2 ...
##  $ ConcValCombo   : Factor w/ 4 levels "abstract-negative",..: 1 4 1 4 1 1 3 3 2 3 ...
##  $ Task           : Factor w/ 2 levels "Concrete","Valence": 2 2 2 2 2 2 2 2 2 2 ...
##  $ BlockOrder     : Factor w/ 2 levels "CV","VC": 2 2 2 2 2 2 2 2 2 2 ...
##  $ Group          : Factor w/ 6 levels "A","abstract;concrete",..: 3 3 3 3 3 3 3 3 3 3 ...
##  $ Response       : Factor w/ 4 levels "abstract","concrete",..: 3 4 3 4 3 3 3 3 4 3 ...
##  $ Accuracy       : Factor w/ 1 level "1": 1 1 1 1 1 1 1 1 1 1 ...
##  $ EventTime      : Factor w/ 11584 levels "1732203168805",..: 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 ...
##  $ Value          : Factor w/ 4 levels "abstract","concrete",..: 3 4 3 4 3 3 3 3 4 3 ...
##  $ RT             : Factor w/ 1691 levels "1270","1272.16666666667",..: 1184 1691 1380 1653 1444 1532 1364 1357 1166 1187 ...
##  $ ReactionTime   : num [1:11585] 1164 1184 1660 1157 1264 ...
##  $ Key_value_F    : Factor w/ 4 levels "abstract","concrete",..: 3 3 3 3 3 3 3 3 3 3 ...
##  $ Key_value_J    : Factor w/ 2 levels "A","B": 2 2 2 2 2 2 2 2 2 2 ...
##  $ Comments       : Factor w/ 0 levels: NA NA NA NA NA NA NA NA NA NA ...
##  $ LogReactionTime: num [1:11585] 7.06 7.08 7.41 7.05 7.14 ...
##  $ LogRT          : Factor w/ 1691 levels "7.14677217945264",..: 1184 1691 1380 1653 1444 1532 1364 1357 1166 1187 ...
##  $ TrialNumber    : Factor w/ 240 levels "1","2","3","4",..: 1 3 4 5 6 7 8 9 10 11 ...
##  $ Category       : Factor w/ 3 levels "Adjs","Nouns",..: 3 3 3 3 3 3 3 3 3 3 ...

conc <- df_factors %>% 
  filter(Task == "Concrete") %>% 
  mutate(Response.n = as.numeric(factor(Response, levels = c("abstract", "concrete"))) - 1,
         Concrete = ifelse(grepl("abstract", ConcValCombo), "abstract", 
                    ifelse(grepl("concrete", ConcValCombo), "concrete", NA)),
        cConcValCombo = as.numeric(ConcValCombo) - mean(as.numeric(ConcValCombo)),
        cConcrete = as.numeric(as.factor(Concrete)) - mean(as.numeric(as.factor(Concrete))),
        # cSyntactic = as.numeric(factor(Syntactic)) - mean(as.numeric(factor(Syntactic)))
  )

m = lmer(LogReactionTime ~ cConcValCombo + (1+cConcValCombo|ID.true) + (1+cConcValCombo|Word), data=conc)
summary(m)

## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: LogReactionTime ~ cConcValCombo + (1 + cConcValCombo | ID.true) +  
##     (1 + cConcValCombo | Word)
##    Data: conc
## 
## REML criterion at convergence: 3137.2
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -3.6179 -0.6346 -0.1791  0.4717  4.8591 
## 
## Random effects:
##  Groups   Name          Variance Std.Dev. Corr 
##  Word     (Intercept)   0.002293 0.04788       
##           cConcValCombo 0.002611 0.05110  0.85 
##  ID.true  (Intercept)   0.043548 0.20868       
##           cConcValCombo 0.003030 0.05504  -0.19
##  Residual               0.094883 0.30803       
## Number of obs: 5624, groups:  Word, 117; ID.true, 52
## 
## Fixed effects:
##                 Estimate Std. Error         df t value Pr(>|t|)    
## (Intercept)    6.7908831  0.0298862 55.2507060 227.224   <2e-16 ***
## cConcValCombo -0.0007486  0.0106036 81.8385078  -0.071    0.944    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##             (Intr)
## cConcValCmb -0.051

No effect of word concreteness

m = lmer(LogReactionTime ~ cConcrete + (1+cConcrete|ID.true) + (1|Word), data=conc)
summary(m)

## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: LogReactionTime ~ cConcrete + (1 + cConcrete | ID.true) + (1 |  
##     Word)
##    Data: conc
## 
## REML criterion at convergence: 3118.4
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -3.6385 -0.6448 -0.1755  0.4805  4.9471 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev. Corr 
##  Word     (Intercept) 0.004746 0.06889       
##  ID.true  (Intercept) 0.043091 0.20758       
##           cConcrete   0.016485 0.12839  -0.18
##  Residual             0.094308 0.30710       
## Number of obs: 5624, groups:  Word, 117; ID.true, 52
## 
## Fixed effects:
##             Estimate Std. Error       df t value Pr(>|t|)    
## (Intercept)  6.78218    0.02979 55.77945 227.655   <2e-16 ***
## cConcrete   -0.02065    0.02349 83.70034  -0.879    0.382    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##           (Intr)
## cConcrete -0.136

Small effect of response choice

m = lmer(LogReactionTime ~ Response.n + (1|ID.true) + (1|Word), data=conc)
summary(m)

## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: LogReactionTime ~ Response.n + (1 | ID.true) + (1 | Word)
##    Data: conc
## 
## REML criterion at convergence: 3239.9
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -3.3853 -0.6470 -0.1793  0.4899  4.9199 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev.
##  Word     (Intercept) 0.005636 0.07507 
##  ID.true  (Intercept) 0.042882 0.20708 
##  Residual             0.097689 0.31255 
## Number of obs: 5624, groups:  Word, 117; ID.true, 52
## 
## Fixed effects:
##              Estimate Std. Error        df t value Pr(>|t|)    
## (Intercept)   6.78896    0.03091  64.64002 219.613   <2e-16 ***
## Response.n   -0.01904    0.01631 103.90862  -1.168    0.246    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##            (Intr)
## Response.n -0.258

In the Valence task , is there a difference between positive and negative on ReactionTime?

Nope.

val <- df.outliers.removed %>% 
  filter(Task == "Valence") %>%
  filter(!is.na(ConcValCombo)) %>% 
  mutate(Response.n = as.numeric(factor(Response, levels = c("negative", "positive"))) - 1,
         Valence = case_when(
                        grepl("negative", ConcValCombo) ~ "negative",
                        grepl("positive", ConcValCombo) ~ "positive",
                        TRUE ~ NA_character_
                    ),
         # Valence = ifelse(grepl("negative", ConcValCombo), "negative", 
         #            ifelse(grepl("positive", ConcValCombo), "positive", NA)),
        cConcValCombo = as.numeric(as.factor(ConcValCombo)) - mean(as.numeric(as.factor(ConcValCombo))),
        cValence = as.numeric(as.factor(Valence)) - mean(as.numeric(as.factor(Valence)))
  )


sum(is.na(val$ConcValCombo))  # Count missing values

## [1] 0

sum(is.na(val$LogReactionTime)) # Check for missing LogReactionTime

## [1] 0

var(val$cConcValCombo)

## [1] 1.245257

unique(val$ConcValCombo)

## [1] "abstract-negative" "concrete-positive" "concrete-negative"
## [4] "abstract-positive"

# View(val)
# valna <- val %>% 
#   filter(is.na(val$ConcValCombo))

m = lmer(LogReactionTime ~ cConcValCombo + (1+cConcValCombo|ID.true) + (1+cConcValCombo|Word), data=val)
summary(m)

## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: LogReactionTime ~ cConcValCombo + (1 + cConcValCombo | ID.true) +  
##     (1 + cConcValCombo | Word)
##    Data: val
## 
## REML criterion at convergence: 1183
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -3.1019 -0.6208 -0.1584  0.4361  5.6861 
## 
## Random effects:
##  Groups   Name          Variance  Std.Dev. Corr 
##  Word     (Intercept)   3.717e-03 0.060968      
##           cConcValCombo 2.332e-05 0.004829 -1.00
##  ID.true  (Intercept)   4.033e-02 0.200812      
##           cConcValCombo 3.050e-04 0.017463 0.08 
##  Residual               6.666e-02 0.258187      
## Number of obs: 5961, groups:  Word, 117; ID.true, 52
## 
## Fixed effects:
##                 Estimate Std. Error         df t value Pr(>|t|)    
## (Intercept)     6.621582   0.028628  55.042064 231.295   <2e-16 ***
## cConcValCombo  -0.005987   0.006351 105.906807  -0.943    0.348    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##             (Intr)
## cConcValCmb 0.002

Effect of Word Valence

m = lmer(LogReactionTime ~ cValence + (1+cValence|ID.true) + (1|Word), data=val)
summary(m)

## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: LogReactionTime ~ cValence + (1 + cValence | ID.true) + (1 |  
##     Word)
##    Data: val
## 
## REML criterion at convergence: 1148.7
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -3.2866 -0.6132 -0.1493  0.4357  5.7974 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev. Corr
##  Word     (Intercept) 0.003815 0.06177      
##  ID.true  (Intercept) 0.040457 0.20114      
##           cValence    0.004277 0.06540  0.11
##  Residual             0.065952 0.25681      
## Number of obs: 5961, groups:  Word, 117; ID.true, 52
## 
## Fixed effects:
##              Estimate Std. Error        df t value Pr(>|t|)    
## (Intercept)   6.62209    0.02868  55.10568 230.882   <2e-16 ***
## cValence     -0.02197    0.01610 108.55084  -1.365    0.175    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##          (Intr)
## cValence 0.063

Effect of response choice

m = lmer(LogReactionTime ~ Response.n + (1|ID.true) + (1|Word), data=val)
summary(m)

## Linear mixed model fit by REML. t-tests use Satterthwaite's method [
## lmerModLmerTest]
## Formula: LogReactionTime ~ Response.n + (1 | ID.true) + (1 | Word)
##    Data: val
## 
## REML criterion at convergence: 1187.9
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -3.1176 -0.6150 -0.1554  0.4411  5.7484 
## 
## Random effects:
##  Groups   Name        Variance Std.Dev.
##  Word     (Intercept) 0.00372  0.06099 
##  ID.true  (Intercept) 0.04037  0.20093 
##  Residual             0.06702  0.25887 
## Number of obs: 5961, groups:  Word, 117; ID.true, 52
## 
## Fixed effects:
##              Estimate Std. Error        df t value Pr(>|t|)    
## (Intercept)   6.63269    0.02938  60.65279 225.759   <2e-16 ***
## Response.n   -0.02209    0.01318 111.12508  -1.676   0.0965 .  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##            (Intr)
## Response.n -0.223

Concret/Abstract Dimension for Verbs,Nouns,Adjs

morgan moyer

2025-02-26

Summary Stats

Start with Accuracy

In these most accurate words, is there still our effect?

Look within categorieslexical

Adjectives

Verbs

Nouns

By Participant

ReactionTime by Task

By Participant

Looking at Responses Generally

Proportion Concrete

Reaction Time analyses

Data Cleaning

Summary Stats

LogReactionTime by Task

ReactionTime by Task

Prop Conc/Val

Does REsponse choice influence RT??

Valence

Concrete

Category

ReactionTime by Task

Look at cases where RT Conc <= RT Valence or Conc Acc >= Val Acc

Conc Acc is higher than Val Acc

Analysis

Is there a difference between Semantic and Valence Tasks?

In the Concreteness task, is there a difference between concreteness and abstractness on ReactionTime?

In the Valence task , is there a difference between positive and negative on ReactionTime?